Simulation Study

Overview

An IaaS service provider in cloud computing offers virtualized computing resources over the internet. They allocate resources to the customers based upon the requests they receive. For large installations, as the resources are available in abundance, immediate allocation of resource is possible. These installations will also have long periods with idle resources, which are occupied using additional strategies, like listing on a spot market. However, these strategies do not suit the operations of small and mid-size providers.

In this project, we focus our study on small and mid-size cloud service providers. We aim to study how various scheduling policies in existence affect the resource allocation in these installations, through simulations. We also aim to achieve a balance between maximum utilization of request, while satisfying large fraction of the customer base.

Current work

One of the prevalent allocation policy in large installations is Load Balancing. It is implemented by simply allocating a task as soon as the required resource becomes available. However, in small and mid-size installations, as the available resources are significantly small compared to the large ones, whether or not load balancing is still a superior option is yet to be proven. Hence, we consider algorithms, which follow biased loading structure and which are proven theoretically to be optimal along with load balancing approach and study the outcomes.

In our simulations, we also aim to keep our input data, as real as possible. Hence, we use existing traces provided by cloud service providers. Even though, the provided traces do not contain every detail regarding the job, it is sufficient to provide us with details required for generating a base trace. We derive the missing details in the trace, by employing statistical methods.

Finally, we also focus on various methods in existence to compare and present results, like Heat maps, and develop them further to help us understand the performance comparison between various allocation policies in a comprehensible manner.

Our repository